EW JOURNAL

A preschooler at the Bing Nursery School on the Stanford campus, somewhere in the late 1960s, has been shown a marshmallow on a plate and given an arrangement she can repeat back. The adult is going to leave the room. If she waits until he returns, she gets two. If she cannot wait, there is a small bell on the table, and ringing it brings him back at once, and then she gets one. He leaves. The door closes. What the camera records over the following fifteen minutes is a person of four or five conducting an experiment on herself with no training and no vocabulary for what she is doing: she covers her eyes, she sits on her own hands, she turns her chair to face the wall, she sings, she talks to the empty air, and at some point she may stumble onto the advanced and slightly unsettling insight that she herself, not the sweet sitting six inches away, is the one who cannot be trusted.

Walter Mischel had come to these studies from a question about what self-control actually consists of, and the answer his group arrived at was less flattering to willpower than the legend that followed. The children who lasted had simply found a way to stop attending to the object: they redescribed the marshmallow as a cloud, looked elsewhere, or filled the interval with something of their own. Attention was the mechanism. Strength of character, in any sense a moralist would recognise, barely entered into it.

What made the study famous outside psychology was the follow-up work, and it is worth being precise about what that follow-up actually claimed. Mischel with Yuichi Shoda and Philip Peake tracked a subset of the original children into adolescence and reported that longer waiting times at four or five went with better academic outcomes and higher SAT scores years later. That is a correlation in a modest sample drawn from the children of Stanford staff and students, a population already skewed toward the household stability and parental education that predict good outcomes on their own. A whole cottage industry of self-help sermonising grew out of the finding anyway, and for years afterward, Mischel kept pushing back against the oversimplified moral, mostly without success. A tidy parable spreads further than a hedged footnote ever will, and he was the one who had written all the hedges.

Two later findings are the ones that change the shape of the thing, and only one of them is well known. The well-known one flatters the test. The other does not, which is presumably why it stayed obscure.

In 2013 Celeste Kidd and colleagues at Rochester ran the marshmallow task on twenty-eight children after first manipulating whether the adult in the room had kept a promise. Some children experienced a grown-up who said he would come back with better art supplies and did. Others experienced a grown-up who said the same thing and returned with nothing. The gap this produced in how long the children waited was not subtle: something like twelve minutes in the reliable condition against something like three in the broken-promise one, a fourfold difference produced by nothing except one prior experience of being told the truth or not. Children on the losing end of a broken promise reached for the marshmallow far sooner, and the authors titled the paper Rational snacking, which is the correct reading and a small masterpiece of understatement. A child who has just watched an adult’s word fail is pricing counterparty risk with the only data available to her when she takes the certain sweet, and pricing it correctly.

Then in 2018 Tyler Watts, Greg Duncan and Haonan Quan attempted a conceptual replication in a much larger and more socioeconomically varied sample, and their result is the one that should have been the headline. The bivariate correlation with later achievement survived. It was about half the size of the original, and it shrank by roughly two thirds once they controlled for family background and early cognitive ability. Something is still there; it is a good deal smaller than the parable requires, and a substantial part of what looked like the child’s virtue turns out to have been the child’s household.

That original experiment is no less remarkable for any of this. What changes is what it is actually an experiment about, once the retelling’s gloss comes off.

Set beside this the longer literature that people often confuse with it. The Dunedin study followed a full birth cohort of just over a thousand New Zealand children from birth into their thirties, measuring behavioural regulation across childhood instead of in a single fifteen-minute test, and Moffitt, Caspi and their colleagues reported a gradient: more self-control in childhood, better adult health, finances and criminal record, with the associations persisting after adjustment for social class and IQ, though reduced, a pattern epidemiologists sometimes call an attenuated but durable effect, which is a duller way of saying the finding mostly survives being checked. That is a different instrument measuring a broader trait over many years, and it holds up considerably better than the marshmallow does. The children in question were born within a single year, April 1972 to March 1973, in one New Zealand city, and the study has kept finding the overwhelming majority of them at every follow-up since, an unusually complete return rate that is itself part of why the findings carry more weight than a typical longitudinal study’s. The distinction matters precisely because the marshmallow is the image everybody has and Dunedin is the evidence everybody cites it for.

Nothing about this stopped the marshmallow test from becoming a fixture of early-childhood policy arguments, cited in support of programmes aimed at teaching poor children self-control as though the deficit, if it existed, began inside the child rather than inside the child’s circumstances. The Rochester finding suggests the arrow may run the other way: teach the world to keep its promises and the patience tends to follow on its own.

Investors receive the identical offer with the charm removed and the term extended. The compounding chart appears in every prospectus, and it convinces on sight, because the arithmetic really is beyond argument: a single marshmallow today was never going to beat two tomorrow, and every child in that room had already worked that out for herself. What the chart cannot draw is the room itself, stretched out across years and lived one trading day at a time, in the presence of a price that moves against you for reasons nobody can explain until afterwards, and with a set of obligations that do not pause while you are being patient.

Here the mechanism is worth stating exactly, since this is where the parable does real damage. What actually kills a long-horizon strategy, far more often than impatience does, is a liability arriving inside the holding period: a margin call, a redemption, a payment date, a lost job, any of which converts a paper drawdown into a realised loss at the worst available price. Two investors holding identical assets with identical temperaments can meet entirely different fates on the basis of whether either of them is forced to sell in a bad month, and nothing about their patience enters that calculation. The compounding curve assumes the position is held. Whether it can be held is a question about the liability side of the household, and the chart has no column for it. The good advice, as a result, is rarely about character at all: keep cash, avoid leverage you cannot service, match the term of the money to the term of the plan. Dull instructions, and they do the work that willpower is credited with.

And the other half, the half the Rochester children understood. Waiting is a bet on the reliability of the counterparty, and the counterparty here is a whole apparatus: currencies, courts, custodians, the assumption that a contract signed today will be honoured in twenty years by institutions not yet staffed. Most of the time, in most of the places where such charts are printed, that bet has paid. It has not paid everywhere, and anyone whose grandparents held government bonds through a currency reform will hold a different view of the second marshmallow’s inevitability, and hold it for reasons that are historical, not psychological. Argentina has defaulted on its sovereign debt repeatedly since independence, most recently in 2020, and an Argentine saver’s doubts about whether tomorrow’s marshmallow arrives on schedule are a bet priced correctly against a long and specific memory, not irrational pessimism.

The mathematics favours the patient. It says nothing whatever about whether the future has agreed to show up, and the five-year-old who checked first was doing something more sophisticated than we gave her credit for.